* SBCL can't cancel computations of futures that are no longer
  referenced and should be garbage-collectable (apparent reason: SBCL
  keeps references to future object somewhere on the stack even when
  using weak-pointer-value like in make-scheduler-task)

* CCL crashes on consing-intensive parallel tasks (for example, values
  >50 on stress test)

* Some of the tests result in one thread pool thread somehow getting
  stuck (check *free-threads* after running the test suite; also this
  causes thread-pool-size-advise test to fail on subsequent test suite
  invocations)

